Who is responsible when AI makes a decision?
The short answer
The people and institutions choosing, deploying, and relying on an AI system still have responsibilities. Maxim’s essays argue that delegating analysis can make responsibility feel dispersed even while real people bear the consequences. Meaningful oversight requires someone who understands the purpose, can question the recommendation, and has the authority to change what happens next.
Responsibility begins to feel distributed, even when it is not.
Someone chooses what the system is for
Before an AI recommendation appears, people have already chosen a goal and a way to measure success. A customer-service process might prioritize speed, cost, or resolving a person’s problem. Those choices shape what the system treats as a good outcome.
The Responsibility We Are Quietly Delegating to Machines puts this work at the center of accountability. Someone must decide what should be optimized, which risks are acceptable, and where the boundaries lie. A system’s ability to pursue an objective does not settle whether the objective is appropriate.
From The Responsibility We Are Quietly Delegating to Machines
Oversight needs time, understanding, and authority
A person can be present in a process while having little practical influence over it. If they see a recommendation only at the last moment, lack the context behind it, or cannot change it, approval can become a routine confirmation rather than an examination.
Maxim’s essay asks what happens as checking gradually feels too expensive. Review needs a clear owner and a real opportunity to interrupt the process. The useful question is what the reviewer can do when the recommendation appears wrong, not simply whether a person clicked a button.
From The Responsibility We Are Quietly Delegating to Machines
Notice the decisions made before the recommendation
Did You Notice How Much Is Decided Before We Arrive? follows a subtler form of delegation: accepting defaults and pre-arranged options. By the time a choice is offered, an interface may already have determined which possibilities are visible and which seem easiest.
Accountability therefore includes the design around an AI response. Who selected the information? Which alternatives were left out? Can the affected person ask for another route? Examining only the final answer can miss decisions that shaped it much earlier.
Make it possible to challenge the outcome
Consider an automated customer-service result that leaves a person’s problem unresolved. Someone needs to be able to reopen the case, understand what happened, and decide whether the process should change. Responsibility becomes practical through that ability to respond to consequences.
These essays offer an ethical and organizational argument about keeping ownership of decisions. They ask readers to identify who can act when a system fails. A useful discussion ends with named responsibilities and a route for review, rather than the explanation that the algorithm recommended it.
From The Responsibility We Are Quietly Delegating to Machines · Did You Notice How Much Is Decided Before We Arrive?
Read the original essays
-
The Responsibility We Are Quietly Delegating to Machines
Explores how useful automation can gradually become unexamined delegation of judgment.
-
Did You Notice How Much Is Decided Before We Arrive?
Shows how defaults and pre-arranged options shape the choices people appear to make.